P5 Medicine and Slowfood AI
Data Science and Mental Health
P5 Medicine and Slowfood AI: Data Science and Mental Health
Magnus Boman and Sumithra Velupillai
Over the last three and a half years or so, we planned for and realised a series of four extremely open-ended roundtables on data science and mental health. Mostly as a gift to our generous participants we wrote up a summary. In it, you will see why our participants thought precision medicine should be defined not only as P4 (personalised, predictive, preventive and participatory) modern medicine, but as P5, adding ‘psychological’ to the list. To entice you to read it, here is a short background.
The first roundtable R1 at the Alan Turing Institute in London back in November 2017, was not conceived of as the start of a series, but as a meeting of minds, without much agenda or presentations. What we all brought was our common interest in sharing thoughts and ideas in an emerging field of research. We were all gathered at one table, a format reused at the subsequent roundtables, with researchers from Sweden and the UK. At this and the three that followed in Stockholm and London, we used an analogue format: no laptops, no slides. We ended up with vivid knowledge and opinion exchanges and discussions, each time under a broad theme, with a literary foreground as inspiration. The analogue format helped open for freer thinking about where this field might head.
The theme for R2 was Transfer Learning, as in the training for one patient population being generalised and possibly adapted to apply to another, but also as in the transfer of learning between scientific disciplines. The former is recognised as one of the harder problems of long-term learning machines, even for narrow domains. We speak of it as ‘slowfood AI.’ The latter is recognised as a challenge when working in interdisciplinary areas, but also as an opportunity in making scientific progress and opening up for new research agendas. The titles (and themes) of Charles Dickens’ novels were used as session ice-breakers.
Methods for Precision Mental Health was the theme for R3, with Selma Lagerlöf as our literary discussion guide. This theme was motivated by the immaturity in theory as well as practice of data science in multimodal and dynamic contexts. In particular, we looked at multimodal input to learning machines where genetic data and/or brain images were combined with text generated during treatment for mental health conditions.
Finally, R4 was organised with a Tangibles theme, each session headed under a Tove Jansson story title. The two of us felt, with some of our most active attendants, that the time was right for looking at tangible output, impact, and usefulness in practice of the four roundtables. Our write-up is one such tangible. We hope you find it useful.
